Chinese Handwriting Style Recognition using Morphological Ensemble DenseNet

Chinese handwriting comes in many different styles. Recognizing them is an important part of studying historical Chinese manuscripts. To help this, we propose using the DenseNet-169 architecture to recognize different writers. We also propose a method to recognize Chinese handwriting styles using an ensemble classifier consisting of three DenseNet-169 models. On two of the models, morphological operations to the images are applied to extract nonlinear features from the input image. Using a soft-voting ensemble mechanism, our method did not outperform the DenseNet model trained on the entire dataset, but we find that DenseNet manages to outperform both ResNet50 and ResNet50. We conclude that the unmodified DenseNet-169 model is better for recognizing Chinese handwriting styles.